Abstract: Continuous wave (CW) telegraph is a crucial communication means for high-frequency tactical communication in emergencies. But there exists serious decline in high-frequency channel, thus the statistical properties of interference noise can not be known in advance. A new adaptive Kalman filter based on autoregressive moving average (ARMA) innovation model is proposed in this paper to detect weak high-frequency CW signal with unknown precise statistical variance of Gaussian noise in system. The state space random signal model of CW signal is firstly defined, by which the ARMA innovation model is constructed. Then by means of the on-line identification of ARMA model parameters, the Kalman filter gain is estimated to implement the adpative Kalman filtering of CW signal. Simulation studies show this method can dynamically track weak CW signal with unknown variance of Gaussian interference noise.